1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
65
from itertools import izip, chain
70
from zlib import Z_DEFAULT_COMPRESSION
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
92
from bzrlib.errors import (
98
KnitDataStreamIncompatible,
101
RevisionAlreadyPresent,
103
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
104
from bzrlib.osutils import (
109
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
110
from bzrlib.tsort import topo_sort
113
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
116
# TODO: Split out code specific to this format into an associated object.
118
# TODO: Can we put in some kind of value to check that the index and data
119
# files belong together?
121
# TODO: accommodate binaries, perhaps by storing a byte count
123
# TODO: function to check whole file
125
# TODO: atomically append data, then measure backwards from the cursor
126
# position after writing to work out where it was located. we may need to
127
# bypass python file buffering.
129
DATA_SUFFIX = '.knit'
130
INDEX_SUFFIX = '.kndx'
133
class KnitContent(object):
134
"""Content of a knit version to which deltas can be applied."""
137
"""Return a list of (origin, text) tuples."""
138
return list(self.annotate_iter())
140
def line_delta_iter(self, new_lines):
141
"""Generate line-based delta from this content to new_lines."""
142
new_texts = new_lines.text()
143
old_texts = self.text()
144
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
145
for tag, i1, i2, j1, j2 in s.get_opcodes():
148
# ofrom, oto, length, data
149
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
151
def line_delta(self, new_lines):
152
return list(self.line_delta_iter(new_lines))
155
def get_line_delta_blocks(knit_delta, source, target):
156
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
157
target_len = len(target)
160
for s_begin, s_end, t_len, new_text in knit_delta:
161
true_n = s_begin - s_pos
164
# knit deltas do not provide reliable info about whether the
165
# last line of a file matches, due to eol handling.
166
if source[s_pos + n -1] != target[t_pos + n -1]:
169
yield s_pos, t_pos, n
170
t_pos += t_len + true_n
172
n = target_len - t_pos
174
if source[s_pos + n -1] != target[t_pos + n -1]:
177
yield s_pos, t_pos, n
178
yield s_pos + (target_len - t_pos), target_len, 0
181
class AnnotatedKnitContent(KnitContent):
182
"""Annotated content."""
184
def __init__(self, lines):
187
def annotate_iter(self):
188
"""Yield tuples of (origin, text) for each content line."""
189
return iter(self._lines)
191
def strip_last_line_newline(self):
192
line = self._lines[-1][1].rstrip('\n')
193
self._lines[-1] = (self._lines[-1][0], line)
196
return [text for origin, text in self._lines]
199
return AnnotatedKnitContent(self._lines[:])
202
class PlainKnitContent(KnitContent):
203
"""Unannotated content.
205
When annotate[_iter] is called on this content, the same version is reported
206
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
210
def __init__(self, lines, version_id):
212
self._version_id = version_id
214
def annotate_iter(self):
215
"""Yield tuples of (origin, text) for each content line."""
216
for line in self._lines:
217
yield self._version_id, line
220
return PlainKnitContent(self._lines[:], self._version_id)
222
def strip_last_line_newline(self):
223
self._lines[-1] = self._lines[-1].rstrip('\n')
229
class KnitAnnotateFactory(object):
230
"""Factory for creating annotated Content objects."""
234
def make(self, lines, version_id):
235
num_lines = len(lines)
236
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
238
def parse_fulltext(self, content, version_id):
239
"""Convert fulltext to internal representation
241
fulltext content is of the format
242
revid(utf8) plaintext\n
243
internal representation is of the format:
246
# TODO: jam 20070209 The tests expect this to be returned as tuples,
247
# but the code itself doesn't really depend on that.
248
# Figure out a way to not require the overhead of turning the
249
# list back into tuples.
250
lines = [tuple(line.split(' ', 1)) for line in content]
251
return AnnotatedKnitContent(lines)
253
def parse_line_delta_iter(self, lines):
254
return iter(self.parse_line_delta(lines))
256
def parse_line_delta(self, lines, version_id):
257
"""Convert a line based delta into internal representation.
259
line delta is in the form of:
260
intstart intend intcount
262
revid(utf8) newline\n
263
internal representation is
264
(start, end, count, [1..count tuples (revid, newline)])
271
def cache_and_return(line):
272
origin, text = line.split(' ', 1)
273
return cache.setdefault(origin, origin), text
275
# walk through the lines parsing.
277
start, end, count = [int(n) for n in header.split(',')]
278
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
279
result.append((start, end, count, contents))
282
def get_fulltext_content(self, lines):
283
"""Extract just the content lines from a fulltext."""
284
return (line.split(' ', 1)[1] for line in lines)
286
def get_linedelta_content(self, lines):
287
"""Extract just the content from a line delta.
289
This doesn't return all of the extra information stored in a delta.
290
Only the actual content lines.
295
header = header.split(',')
296
count = int(header[2])
297
for i in xrange(count):
298
origin, text = next().split(' ', 1)
301
def lower_fulltext(self, content):
302
"""convert a fulltext content record into a serializable form.
304
see parse_fulltext which this inverts.
306
# TODO: jam 20070209 We only do the caching thing to make sure that
307
# the origin is a valid utf-8 line, eventually we could remove it
308
return ['%s %s' % (o, t) for o, t in content._lines]
310
def lower_line_delta(self, delta):
311
"""convert a delta into a serializable form.
313
See parse_line_delta which this inverts.
315
# TODO: jam 20070209 We only do the caching thing to make sure that
316
# the origin is a valid utf-8 line, eventually we could remove it
318
for start, end, c, lines in delta:
319
out.append('%d,%d,%d\n' % (start, end, c))
320
out.extend(origin + ' ' + text
321
for origin, text in lines)
324
def annotate_iter(self, knit, version_id):
325
content = knit._get_content(version_id)
326
return content.annotate_iter()
329
class KnitPlainFactory(object):
330
"""Factory for creating plain Content objects."""
334
def make(self, lines, version_id):
335
return PlainKnitContent(lines, version_id)
337
def parse_fulltext(self, content, version_id):
338
"""This parses an unannotated fulltext.
340
Note that this is not a noop - the internal representation
341
has (versionid, line) - its just a constant versionid.
343
return self.make(content, version_id)
345
def parse_line_delta_iter(self, lines, version_id):
347
num_lines = len(lines)
348
while cur < num_lines:
351
start, end, c = [int(n) for n in header.split(',')]
352
yield start, end, c, lines[cur:cur+c]
355
def parse_line_delta(self, lines, version_id):
356
return list(self.parse_line_delta_iter(lines, version_id))
358
def get_fulltext_content(self, lines):
359
"""Extract just the content lines from a fulltext."""
362
def get_linedelta_content(self, lines):
363
"""Extract just the content from a line delta.
365
This doesn't return all of the extra information stored in a delta.
366
Only the actual content lines.
371
header = header.split(',')
372
count = int(header[2])
373
for i in xrange(count):
376
def lower_fulltext(self, content):
377
return content.text()
379
def lower_line_delta(self, delta):
381
for start, end, c, lines in delta:
382
out.append('%d,%d,%d\n' % (start, end, c))
386
def annotate_iter(self, knit, version_id):
387
return annotate_knit(knit, version_id)
390
def make_empty_knit(transport, relpath):
391
"""Construct a empty knit at the specified location."""
392
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
395
class KnitVersionedFile(VersionedFile):
396
"""Weave-like structure with faster random access.
398
A knit stores a number of texts and a summary of the relationships
399
between them. Texts are identified by a string version-id. Texts
400
are normally stored and retrieved as a series of lines, but can
401
also be passed as single strings.
403
Lines are stored with the trailing newline (if any) included, to
404
avoid special cases for files with no final newline. Lines are
405
composed of 8-bit characters, not unicode. The combination of
406
these approaches should mean any 'binary' file can be safely
407
stored and retrieved.
410
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
411
factory=None, delta=True, create=False, create_parent_dir=False,
412
delay_create=False, dir_mode=None, index=None, access_method=None):
413
"""Construct a knit at location specified by relpath.
415
:param create: If not True, only open an existing knit.
416
:param create_parent_dir: If True, create the parent directory if
417
creating the file fails. (This is used for stores with
418
hash-prefixes that may not exist yet)
419
:param delay_create: The calling code is aware that the knit won't
420
actually be created until the first data is stored.
421
:param index: An index to use for the knit.
423
if access_mode is None:
425
super(KnitVersionedFile, self).__init__(access_mode)
426
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
427
self.transport = transport
428
self.filename = relpath
429
self.factory = factory or KnitAnnotateFactory()
430
self.writable = (access_mode == 'w')
433
self._max_delta_chain = 200
436
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
437
access_mode, create=create, file_mode=file_mode,
438
create_parent_dir=create_parent_dir, delay_create=delay_create,
442
if access_method is None:
443
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
444
((create and not len(self)) and delay_create), create_parent_dir)
446
_access = access_method
447
if create and not len(self) and not delay_create:
449
self._data = _KnitData(_access)
452
return '%s(%s)' % (self.__class__.__name__,
453
self.transport.abspath(self.filename))
455
def _check_should_delta(self, first_parents):
456
"""Iterate back through the parent listing, looking for a fulltext.
458
This is used when we want to decide whether to add a delta or a new
459
fulltext. It searches for _max_delta_chain parents. When it finds a
460
fulltext parent, it sees if the total size of the deltas leading up to
461
it is large enough to indicate that we want a new full text anyway.
463
Return True if we should create a new delta, False if we should use a
468
delta_parents = first_parents
469
for count in xrange(self._max_delta_chain):
470
parent = delta_parents[0]
471
method = self._index.get_method(parent)
472
index, pos, size = self._index.get_position(parent)
473
if method == 'fulltext':
477
delta_parents = self._index.get_parents(parent)
479
# We couldn't find a fulltext, so we must create a new one
482
return fulltext_size > delta_size
484
def _add_raw_records(self, records, data):
485
"""Add all the records 'records' with data pre-joined in 'data'.
487
:param records: A list of tuples(version_id, options, parents, size).
488
:param data: The data for the records. When it is written, the records
489
are adjusted to have pos pointing into data by the sum of
490
the preceding records sizes.
493
raw_record_sizes = [record[3] for record in records]
494
positions = self._data.add_raw_records(raw_record_sizes, data)
497
for (version_id, options, parents, size), access_memo in zip(
499
index_entries.append((version_id, options, access_memo, parents))
500
if self._data._do_cache:
501
self._data._cache[version_id] = data[offset:offset+size]
503
self._index.add_versions(index_entries)
505
def enable_cache(self):
506
"""Start caching data for this knit"""
507
self._data.enable_cache()
509
def clear_cache(self):
510
"""Clear the data cache only."""
511
self._data.clear_cache()
513
def copy_to(self, name, transport):
514
"""See VersionedFile.copy_to()."""
515
# copy the current index to a temp index to avoid racing with local
517
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
518
self.transport.get(self._index._filename))
520
f = self._data._open_file()
522
transport.put_file(name + DATA_SUFFIX, f)
525
# move the copied index into place
526
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
528
def create_empty(self, name, transport, mode=None):
529
return KnitVersionedFile(name, transport, factory=self.factory,
530
delta=self.delta, create=True)
532
def _fix_parents(self, version_id, new_parents):
533
"""Fix the parents list for version.
535
This is done by appending a new version to the index
536
with identical data except for the parents list.
537
the parents list must be a superset of the current
540
current_values = self._index._cache[version_id]
541
assert set(current_values[4]).difference(set(new_parents)) == set()
542
self._index.add_version(version_id,
544
(None, current_values[2], current_values[3]),
547
def get_data_stream(self, required_versions):
548
"""Get a data stream for the specified versions.
550
Versions may be returned in any order, not necessarily the order
553
:param required_versions: The exact set of versions to be extracted.
554
Unlike some other knit methods, this is not used to generate a
555
transitive closure, rather it is used precisely as given.
557
:returns: format_signature, list of (version, options, length, parents),
560
required_versions = set([osutils.safe_revision_id(v) for v in
562
# we don't care about inclusions, the caller cares.
563
# but we need to setup a list of records to visit.
564
for version_id in required_versions:
565
if not self.has_version(version_id):
566
raise RevisionNotPresent(version_id, self.filename)
567
# Pick the desired versions out of the index in oldest-to-newest order
569
for version_id in self.versions():
570
if version_id in required_versions:
571
version_list.append(version_id)
573
# create the list of version information for the result
574
copy_queue_records = []
576
result_version_list = []
577
for version_id in version_list:
578
options = self._index.get_options(version_id)
579
parents = self._index.get_parents_with_ghosts(version_id)
580
index_memo = self._index.get_position(version_id)
581
copy_queue_records.append((version_id, index_memo))
582
none, data_pos, data_size = index_memo
583
copy_set.add(version_id)
584
# version, options, length, parents
585
result_version_list.append((version_id, options, data_size,
588
# Read the compressed record data.
590
# From here down to the return should really be logic in the returned
591
# callable -- in a class that adapts read_records_iter_raw to read
594
for (version_id, raw_data), \
595
(version_id2, options, _, parents) in \
596
izip(self._data.read_records_iter_raw(copy_queue_records),
597
result_version_list):
598
assert version_id == version_id2, 'logic error, inconsistent results'
599
raw_datum.append(raw_data)
600
pseudo_file = StringIO(''.join(raw_datum))
603
return pseudo_file.read()
605
return pseudo_file.read(length)
606
return (self.get_format_signature(), result_version_list, read)
608
def _extract_blocks(self, version_id, source, target):
609
if self._index.get_method(version_id) != 'line-delta':
611
parent, sha1, noeol, delta = self.get_delta(version_id)
612
return KnitContent.get_line_delta_blocks(delta, source, target)
614
def get_delta(self, version_id):
615
"""Get a delta for constructing version from some other version."""
616
version_id = osutils.safe_revision_id(version_id)
617
self.check_not_reserved_id(version_id)
618
parents = self.get_parents(version_id)
623
index_memo = self._index.get_position(version_id)
624
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
625
noeol = 'no-eol' in self._index.get_options(version_id)
626
if 'fulltext' == self._index.get_method(version_id):
627
new_content = self.factory.parse_fulltext(data, version_id)
628
if parent is not None:
629
reference_content = self._get_content(parent)
630
old_texts = reference_content.text()
633
new_texts = new_content.text()
634
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
636
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
638
delta = self.factory.parse_line_delta(data, version_id)
639
return parent, sha1, noeol, delta
641
def get_format_signature(self):
642
"""See VersionedFile.get_format_signature()."""
643
if self.factory.annotated:
644
annotated_part = "annotated"
646
annotated_part = "plain"
647
return "knit-%s" % (annotated_part,)
649
def get_graph_with_ghosts(self):
650
"""See VersionedFile.get_graph_with_ghosts()."""
651
graph_items = self._index.get_graph()
652
return dict(graph_items)
654
def get_sha1(self, version_id):
655
return self.get_sha1s([version_id])[0]
657
def get_sha1s(self, version_ids):
658
"""See VersionedFile.get_sha1()."""
659
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
660
record_map = self._get_record_map(version_ids)
661
# record entry 2 is the 'digest'.
662
return [record_map[v][2] for v in version_ids]
666
"""See VersionedFile.get_suffixes()."""
667
return [DATA_SUFFIX, INDEX_SUFFIX]
669
def has_ghost(self, version_id):
670
"""True if there is a ghost reference in the file to version_id."""
671
version_id = osutils.safe_revision_id(version_id)
673
if self.has_version(version_id):
675
# optimisable if needed by memoising the _ghosts set.
676
items = self._index.get_graph()
677
for node, parents in items:
678
for parent in parents:
679
if parent not in self._index._cache:
680
if parent == version_id:
684
def insert_data_stream(self, (format, data_list, reader_callable)):
685
"""Insert knit records from a data stream into this knit.
687
If a version in the stream is already present in this knit, it will not
688
be inserted a second time. It will be checked for consistency with the
689
stored version however, and may cause a KnitCorrupt error to be raised
690
if the data in the stream disagrees with the already stored data.
692
:seealso: get_data_stream
694
if format != self.get_format_signature():
695
trace.mutter('incompatible format signature inserting to %r', self)
696
raise KnitDataStreamIncompatible(
697
format, self.get_format_signature())
699
for version_id, options, length, parents in data_list:
700
if self.has_version(version_id):
701
# First check: the list of parents.
702
my_parents = self.get_parents_with_ghosts(version_id)
703
if my_parents != parents:
704
# XXX: KnitCorrupt is not quite the right exception here.
707
'parents list %r from data stream does not match '
708
'already recorded parents %r for %s'
709
% (parents, my_parents, version_id))
711
# Also check the SHA-1 of the fulltext this content will
713
raw_data = reader_callable(length)
714
my_fulltext_sha1 = self.get_sha1(version_id)
715
df, rec = self._data._parse_record_header(version_id, raw_data)
716
stream_fulltext_sha1 = rec[3]
717
if my_fulltext_sha1 != stream_fulltext_sha1:
718
# Actually, we don't know if it's this knit that's corrupt,
719
# or the data stream we're trying to insert.
721
self.filename, 'sha-1 does not match %s' % version_id)
723
self._add_raw_records(
724
[(version_id, options, parents, length)],
725
reader_callable(length))
728
"""See VersionedFile.versions."""
729
if 'evil' in debug.debug_flags:
730
trace.mutter_callsite(2, "versions scales with size of history")
731
return self._index.get_versions()
733
def has_version(self, version_id):
734
"""See VersionedFile.has_version."""
735
if 'evil' in debug.debug_flags:
736
trace.mutter_callsite(2, "has_version is a LBYL scenario")
737
version_id = osutils.safe_revision_id(version_id)
738
return self._index.has_version(version_id)
740
__contains__ = has_version
742
def _merge_annotations(self, content, parents, parent_texts={},
743
delta=None, annotated=None,
744
left_matching_blocks=None):
745
"""Merge annotations for content. This is done by comparing
746
the annotations based on changed to the text.
748
if left_matching_blocks is not None:
749
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
753
for parent_id in parents:
754
merge_content = self._get_content(parent_id, parent_texts)
755
if (parent_id == parents[0] and delta_seq is not None):
758
seq = patiencediff.PatienceSequenceMatcher(
759
None, merge_content.text(), content.text())
760
for i, j, n in seq.get_matching_blocks():
763
# this appears to copy (origin, text) pairs across to the
764
# new content for any line that matches the last-checked
766
content._lines[j:j+n] = merge_content._lines[i:i+n]
768
if delta_seq is None:
769
reference_content = self._get_content(parents[0], parent_texts)
770
new_texts = content.text()
771
old_texts = reference_content.text()
772
delta_seq = patiencediff.PatienceSequenceMatcher(
773
None, old_texts, new_texts)
774
return self._make_line_delta(delta_seq, content)
776
def _make_line_delta(self, delta_seq, new_content):
777
"""Generate a line delta from delta_seq and new_content."""
779
for op in delta_seq.get_opcodes():
782
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
785
def _get_components_positions(self, version_ids):
786
"""Produce a map of position data for the components of versions.
788
This data is intended to be used for retrieving the knit records.
790
A dict of version_id to (method, data_pos, data_size, next) is
792
method is the way referenced data should be applied.
793
data_pos is the position of the data in the knit.
794
data_size is the size of the data in the knit.
795
next is the build-parent of the version, or None for fulltexts.
798
for version_id in version_ids:
801
while cursor is not None and cursor not in component_data:
802
method = self._index.get_method(cursor)
803
if method == 'fulltext':
806
next = self.get_parents(cursor)[0]
807
index_memo = self._index.get_position(cursor)
808
component_data[cursor] = (method, index_memo, next)
810
return component_data
812
def _get_content(self, version_id, parent_texts={}):
813
"""Returns a content object that makes up the specified
815
cached_version = parent_texts.get(version_id, None)
816
if cached_version is not None:
817
if not self.has_version(version_id):
818
raise RevisionNotPresent(version_id, self.filename)
819
return cached_version
821
text_map, contents_map = self._get_content_maps([version_id])
822
return contents_map[version_id]
824
def _check_versions_present(self, version_ids):
825
"""Check that all specified versions are present."""
826
self._index.check_versions_present(version_ids)
828
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
829
nostore_sha, random_id, check_content):
830
"""See VersionedFile.add_lines_with_ghosts()."""
831
self._check_add(version_id, lines, random_id, check_content)
832
return self._add(version_id, lines, parents, self.delta,
833
parent_texts, None, nostore_sha)
835
def _add_lines(self, version_id, parents, lines, parent_texts,
836
left_matching_blocks, nostore_sha, random_id, check_content):
837
"""See VersionedFile.add_lines."""
838
self._check_add(version_id, lines, random_id, check_content)
839
self._check_versions_present(parents)
840
return self._add(version_id, lines[:], parents, self.delta,
841
parent_texts, left_matching_blocks, nostore_sha)
843
def _check_add(self, version_id, lines, random_id, check_content):
844
"""check that version_id and lines are safe to add."""
845
if contains_whitespace(version_id):
846
raise InvalidRevisionId(version_id, self.filename)
847
self.check_not_reserved_id(version_id)
848
# Technically this could be avoided if we are happy to allow duplicate
849
# id insertion when other things than bzr core insert texts, but it
850
# seems useful for folk using the knit api directly to have some safety
851
# blanket that we can disable.
852
if not random_id and self.has_version(version_id):
853
raise RevisionAlreadyPresent(version_id, self.filename)
855
self._check_lines_not_unicode(lines)
856
self._check_lines_are_lines(lines)
858
def _add(self, version_id, lines, parents, delta, parent_texts,
859
left_matching_blocks, nostore_sha):
860
"""Add a set of lines on top of version specified by parents.
862
If delta is true, compress the text as a line-delta against
865
Any versions not present will be converted into ghosts.
867
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
868
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
869
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
870
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
871
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
872
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
873
# +1383 0 8.0370 8.0370 +<len>
874
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
875
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
876
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
877
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
880
if parent_texts is None:
882
for parent in parents:
883
if self.has_version(parent):
884
present_parents.append(parent)
886
# can only compress against the left most present parent.
888
(len(present_parents) == 0 or
889
present_parents[0] != parents[0])):
892
digest = sha_strings(lines)
893
if nostore_sha == digest:
894
raise errors.ExistingContent
895
text_length = sum(map(len, lines))
898
if lines[-1][-1] != '\n':
899
# copy the contents of lines.
901
options.append('no-eol')
902
lines[-1] = lines[-1] + '\n'
905
# To speed the extract of texts the delta chain is limited
906
# to a fixed number of deltas. This should minimize both
907
# I/O and the time spend applying deltas.
908
delta = self._check_should_delta(present_parents)
910
assert isinstance(version_id, str)
911
content = self.factory.make(lines, version_id)
912
if delta or (self.factory.annotated and len(present_parents) > 0):
913
# Merge annotations from parent texts if needed.
914
delta_hunks = self._merge_annotations(content, present_parents,
915
parent_texts, delta, self.factory.annotated,
916
left_matching_blocks)
919
options.append('line-delta')
920
store_lines = self.factory.lower_line_delta(delta_hunks)
922
options.append('fulltext')
923
store_lines = self.factory.lower_fulltext(content)
925
access_memo = self._data.add_record(version_id, digest, store_lines)
926
self._index.add_version(version_id, options, access_memo, parents)
927
return digest, text_length, content
929
def check(self, progress_bar=None):
930
"""See VersionedFile.check()."""
932
def _clone_text(self, new_version_id, old_version_id, parents):
933
"""See VersionedFile.clone_text()."""
934
# FIXME RBC 20060228 make fast by only inserting an index with null
936
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
938
def get_lines(self, version_id):
939
"""See VersionedFile.get_lines()."""
940
return self.get_line_list([version_id])[0]
942
def _get_record_map(self, version_ids):
943
"""Produce a dictionary of knit records.
945
The keys are version_ids, the values are tuples of (method, content,
947
method is the way the content should be applied.
948
content is a KnitContent object.
949
digest is the SHA1 digest of this version id after all steps are done
950
next is the build-parent of the version, i.e. the leftmost ancestor.
951
If the method is fulltext, next will be None.
953
position_map = self._get_components_positions(version_ids)
954
# c = component_id, m = method, i_m = index_memo, n = next
955
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
957
for component_id, content, digest in \
958
self._data.read_records_iter(records):
959
method, index_memo, next = position_map[component_id]
960
record_map[component_id] = method, content, digest, next
964
def get_text(self, version_id):
965
"""See VersionedFile.get_text"""
966
return self.get_texts([version_id])[0]
968
def get_texts(self, version_ids):
969
return [''.join(l) for l in self.get_line_list(version_ids)]
971
def get_line_list(self, version_ids):
972
"""Return the texts of listed versions as a list of strings."""
973
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
974
for version_id in version_ids:
975
self.check_not_reserved_id(version_id)
976
text_map, content_map = self._get_content_maps(version_ids)
977
return [text_map[v] for v in version_ids]
979
_get_lf_split_line_list = get_line_list
981
def _get_content_maps(self, version_ids):
982
"""Produce maps of text and KnitContents
984
:return: (text_map, content_map) where text_map contains the texts for
985
the requested versions and content_map contains the KnitContents.
986
Both dicts take version_ids as their keys.
988
record_map = self._get_record_map(version_ids)
993
for version_id in version_ids:
996
while cursor is not None:
997
method, data, digest, next = record_map[cursor]
998
components.append((cursor, method, data, digest))
999
if cursor in content_map:
1004
for component_id, method, data, digest in reversed(components):
1005
if component_id in content_map:
1006
content = content_map[component_id]
1008
if method == 'fulltext':
1009
assert content is None
1010
content = self.factory.parse_fulltext(data, version_id)
1011
elif method == 'line-delta':
1012
delta = self.factory.parse_line_delta(data, version_id)
1013
content = content.copy()
1014
content._lines = self._apply_delta(content._lines,
1016
content_map[component_id] = content
1018
if 'no-eol' in self._index.get_options(version_id):
1019
content = content.copy()
1020
content.strip_last_line_newline()
1021
final_content[version_id] = content
1023
# digest here is the digest from the last applied component.
1024
text = content.text()
1025
if sha_strings(text) != digest:
1026
raise KnitCorrupt(self.filename,
1027
'sha-1 does not match %s' % version_id)
1029
text_map[version_id] = text
1030
return text_map, final_content
1032
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1034
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1035
if version_ids is None:
1036
version_ids = self.versions()
1038
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1040
pb = progress.DummyProgress()
1041
# we don't care about inclusions, the caller cares.
1042
# but we need to setup a list of records to visit.
1043
# we need version_id, position, length
1044
version_id_records = []
1045
requested_versions = set(version_ids)
1046
# filter for available versions
1047
for version_id in requested_versions:
1048
if not self.has_version(version_id):
1049
raise RevisionNotPresent(version_id, self.filename)
1050
# get a in-component-order queue:
1051
for version_id in self.versions():
1052
if version_id in requested_versions:
1053
index_memo = self._index.get_position(version_id)
1054
version_id_records.append((version_id, index_memo))
1056
total = len(version_id_records)
1057
for version_idx, (version_id, data, sha_value) in \
1058
enumerate(self._data.read_records_iter(version_id_records)):
1059
pb.update('Walking content.', version_idx, total)
1060
method = self._index.get_method(version_id)
1062
assert method in ('fulltext', 'line-delta')
1063
if method == 'fulltext':
1064
line_iterator = self.factory.get_fulltext_content(data)
1066
line_iterator = self.factory.get_linedelta_content(data)
1067
for line in line_iterator:
1070
pb.update('Walking content.', total, total)
1072
def iter_parents(self, version_ids):
1073
"""Iterate through the parents for many version ids.
1075
:param version_ids: An iterable yielding version_ids.
1076
:return: An iterator that yields (version_id, parents). Requested
1077
version_ids not present in the versioned file are simply skipped.
1078
The order is undefined, allowing for different optimisations in
1079
the underlying implementation.
1081
version_ids = [osutils.safe_revision_id(version_id) for
1082
version_id in version_ids]
1083
return self._index.iter_parents(version_ids)
1085
def num_versions(self):
1086
"""See VersionedFile.num_versions()."""
1087
return self._index.num_versions()
1089
__len__ = num_versions
1091
def annotate_iter(self, version_id):
1092
"""See VersionedFile.annotate_iter."""
1093
version_id = osutils.safe_revision_id(version_id)
1094
return self.factory.annotate_iter(self, version_id)
1096
def get_parents(self, version_id):
1097
"""See VersionedFile.get_parents."""
1100
# 52554 calls in 1264 872 internal down from 3674
1101
version_id = osutils.safe_revision_id(version_id)
1103
return self._index.get_parents(version_id)
1105
raise RevisionNotPresent(version_id, self.filename)
1107
def get_parents_with_ghosts(self, version_id):
1108
"""See VersionedFile.get_parents."""
1109
version_id = osutils.safe_revision_id(version_id)
1111
return self._index.get_parents_with_ghosts(version_id)
1113
raise RevisionNotPresent(version_id, self.filename)
1115
def get_ancestry(self, versions, topo_sorted=True):
1116
"""See VersionedFile.get_ancestry."""
1117
if isinstance(versions, basestring):
1118
versions = [versions]
1121
versions = [osutils.safe_revision_id(v) for v in versions]
1122
return self._index.get_ancestry(versions, topo_sorted)
1124
def get_ancestry_with_ghosts(self, versions):
1125
"""See VersionedFile.get_ancestry_with_ghosts."""
1126
if isinstance(versions, basestring):
1127
versions = [versions]
1130
versions = [osutils.safe_revision_id(v) for v in versions]
1131
return self._index.get_ancestry_with_ghosts(versions)
1133
def plan_merge(self, ver_a, ver_b):
1134
"""See VersionedFile.plan_merge."""
1135
ver_a = osutils.safe_revision_id(ver_a)
1136
ver_b = osutils.safe_revision_id(ver_b)
1137
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1139
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1140
annotated_a = self.annotate(ver_a)
1141
annotated_b = self.annotate(ver_b)
1142
return merge._plan_annotate_merge(annotated_a, annotated_b,
1143
ancestors_a, ancestors_b)
1146
class _KnitComponentFile(object):
1147
"""One of the files used to implement a knit database"""
1149
def __init__(self, transport, filename, mode, file_mode=None,
1150
create_parent_dir=False, dir_mode=None):
1151
self._transport = transport
1152
self._filename = filename
1154
self._file_mode = file_mode
1155
self._dir_mode = dir_mode
1156
self._create_parent_dir = create_parent_dir
1157
self._need_to_create = False
1159
def _full_path(self):
1160
"""Return the full path to this file."""
1161
return self._transport.base + self._filename
1163
def check_header(self, fp):
1164
line = fp.readline()
1166
# An empty file can actually be treated as though the file doesn't
1168
raise errors.NoSuchFile(self._full_path())
1169
if line != self.HEADER:
1170
raise KnitHeaderError(badline=line,
1171
filename=self._transport.abspath(self._filename))
1174
return '%s(%s)' % (self.__class__.__name__, self._filename)
1177
class _KnitIndex(_KnitComponentFile):
1178
"""Manages knit index file.
1180
The index is already kept in memory and read on startup, to enable
1181
fast lookups of revision information. The cursor of the index
1182
file is always pointing to the end, making it easy to append
1185
_cache is a cache for fast mapping from version id to a Index
1188
_history is a cache for fast mapping from indexes to version ids.
1190
The index data format is dictionary compressed when it comes to
1191
parent references; a index entry may only have parents that with a
1192
lover index number. As a result, the index is topological sorted.
1194
Duplicate entries may be written to the index for a single version id
1195
if this is done then the latter one completely replaces the former:
1196
this allows updates to correct version and parent information.
1197
Note that the two entries may share the delta, and that successive
1198
annotations and references MUST point to the first entry.
1200
The index file on disc contains a header, followed by one line per knit
1201
record. The same revision can be present in an index file more than once.
1202
The first occurrence gets assigned a sequence number starting from 0.
1204
The format of a single line is
1205
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1206
REVISION_ID is a utf8-encoded revision id
1207
FLAGS is a comma separated list of flags about the record. Values include
1208
no-eol, line-delta, fulltext.
1209
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1210
that the the compressed data starts at.
1211
LENGTH is the ascii representation of the length of the data file.
1212
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1214
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1215
revision id already in the knit that is a parent of REVISION_ID.
1216
The ' :' marker is the end of record marker.
1219
when a write is interrupted to the index file, it will result in a line
1220
that does not end in ' :'. If the ' :' is not present at the end of a line,
1221
or at the end of the file, then the record that is missing it will be
1222
ignored by the parser.
1224
When writing new records to the index file, the data is preceded by '\n'
1225
to ensure that records always start on new lines even if the last write was
1226
interrupted. As a result its normal for the last line in the index to be
1227
missing a trailing newline. One can be added with no harmful effects.
1230
HEADER = "# bzr knit index 8\n"
1232
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1233
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1235
def _cache_version(self, version_id, options, pos, size, parents):
1236
"""Cache a version record in the history array and index cache.
1238
This is inlined into _load_data for performance. KEEP IN SYNC.
1239
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1242
# only want the _history index to reference the 1st index entry
1244
if version_id not in self._cache:
1245
index = len(self._history)
1246
self._history.append(version_id)
1248
index = self._cache[version_id][5]
1249
self._cache[version_id] = (version_id,
1256
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1257
create_parent_dir=False, delay_create=False, dir_mode=None):
1258
_KnitComponentFile.__init__(self, transport, filename, mode,
1259
file_mode=file_mode,
1260
create_parent_dir=create_parent_dir,
1263
# position in _history is the 'official' index for a revision
1264
# but the values may have come from a newer entry.
1265
# so - wc -l of a knit index is != the number of unique names
1269
fp = self._transport.get(self._filename)
1271
# _load_data may raise NoSuchFile if the target knit is
1273
_load_data(self, fp)
1277
if mode != 'w' or not create:
1280
self._need_to_create = True
1282
self._transport.put_bytes_non_atomic(
1283
self._filename, self.HEADER, mode=self._file_mode)
1285
def get_graph(self):
1286
"""Return a list of the node:parents lists from this knit index."""
1287
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1289
def get_ancestry(self, versions, topo_sorted=True):
1290
"""See VersionedFile.get_ancestry."""
1291
# get a graph of all the mentioned versions:
1293
pending = set(versions)
1296
version = pending.pop()
1299
parents = [p for p in cache[version][4] if p in cache]
1301
raise RevisionNotPresent(version, self._filename)
1302
# if not completed and not a ghost
1303
pending.update([p for p in parents if p not in graph])
1304
graph[version] = parents
1307
return topo_sort(graph.items())
1309
def get_ancestry_with_ghosts(self, versions):
1310
"""See VersionedFile.get_ancestry_with_ghosts."""
1311
# get a graph of all the mentioned versions:
1312
self.check_versions_present(versions)
1315
pending = set(versions)
1317
version = pending.pop()
1319
parents = cache[version][4]
1325
pending.update([p for p in parents if p not in graph])
1326
graph[version] = parents
1327
return topo_sort(graph.items())
1329
def iter_parents(self, version_ids):
1330
"""Iterate through the parents for many version ids.
1332
:param version_ids: An iterable yielding version_ids.
1333
:return: An iterator that yields (version_id, parents). Requested
1334
version_ids not present in the versioned file are simply skipped.
1335
The order is undefined, allowing for different optimisations in
1336
the underlying implementation.
1338
for version_id in version_ids:
1340
yield version_id, tuple(self.get_parents(version_id))
1344
def num_versions(self):
1345
return len(self._history)
1347
__len__ = num_versions
1349
def get_versions(self):
1350
"""Get all the versions in the file. not topologically sorted."""
1351
return self._history
1353
def _version_list_to_index(self, versions):
1356
for version in versions:
1357
if version in cache:
1358
# -- inlined lookup() --
1359
result_list.append(str(cache[version][5]))
1360
# -- end lookup () --
1362
result_list.append('.' + version)
1363
return ' '.join(result_list)
1365
def add_version(self, version_id, options, index_memo, parents):
1366
"""Add a version record to the index."""
1367
self.add_versions(((version_id, options, index_memo, parents),))
1369
def add_versions(self, versions):
1370
"""Add multiple versions to the index.
1372
:param versions: a list of tuples:
1373
(version_id, options, pos, size, parents).
1376
orig_history = self._history[:]
1377
orig_cache = self._cache.copy()
1380
for version_id, options, (index, pos, size), parents in versions:
1381
line = "\n%s %s %s %s %s :" % (version_id,
1385
self._version_list_to_index(parents))
1386
assert isinstance(line, str), \
1387
'content must be utf-8 encoded: %r' % (line,)
1389
self._cache_version(version_id, options, pos, size, parents)
1390
if not self._need_to_create:
1391
self._transport.append_bytes(self._filename, ''.join(lines))
1394
sio.write(self.HEADER)
1395
sio.writelines(lines)
1397
self._transport.put_file_non_atomic(self._filename, sio,
1398
create_parent_dir=self._create_parent_dir,
1399
mode=self._file_mode,
1400
dir_mode=self._dir_mode)
1401
self._need_to_create = False
1403
# If any problems happen, restore the original values and re-raise
1404
self._history = orig_history
1405
self._cache = orig_cache
1408
def has_version(self, version_id):
1409
"""True if the version is in the index."""
1410
return version_id in self._cache
1412
def get_position(self, version_id):
1413
"""Return details needed to access the version.
1415
.kndx indices do not support split-out data, so return None for the
1418
:return: a tuple (None, data position, size) to hand to the access
1419
logic to get the record.
1421
entry = self._cache[version_id]
1422
return None, entry[2], entry[3]
1424
def get_method(self, version_id):
1425
"""Return compression method of specified version."""
1427
options = self._cache[version_id][1]
1429
raise RevisionNotPresent(version_id, self._filename)
1430
if 'fulltext' in options:
1433
if 'line-delta' not in options:
1434
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1437
def get_options(self, version_id):
1438
"""Return a string represention options.
1442
return self._cache[version_id][1]
1444
def get_parents(self, version_id):
1445
"""Return parents of specified version ignoring ghosts."""
1446
return [parent for parent in self._cache[version_id][4]
1447
if parent in self._cache]
1449
def get_parents_with_ghosts(self, version_id):
1450
"""Return parents of specified version with ghosts."""
1451
return self._cache[version_id][4]
1453
def check_versions_present(self, version_ids):
1454
"""Check that all specified versions are present."""
1456
for version_id in version_ids:
1457
if version_id not in cache:
1458
raise RevisionNotPresent(version_id, self._filename)
1461
class KnitGraphIndex(object):
1462
"""A knit index that builds on GraphIndex."""
1464
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1465
"""Construct a KnitGraphIndex on a graph_index.
1467
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1468
:param deltas: Allow delta-compressed records.
1469
:param add_callback: If not None, allow additions to the index and call
1470
this callback with a list of added GraphIndex nodes:
1471
[(node, value, node_refs), ...]
1472
:param parents: If True, record knits parents, if not do not record
1475
self._graph_index = graph_index
1476
self._deltas = deltas
1477
self._add_callback = add_callback
1478
self._parents = parents
1479
if deltas and not parents:
1480
raise KnitCorrupt(self, "Cannot do delta compression without "
1483
def _get_entries(self, keys, check_present=False):
1484
"""Get the entries for keys.
1486
:param keys: An iterable of index keys, - 1-tuples.
1491
for node in self._graph_index.iter_entries(keys):
1493
found_keys.add(node[1])
1495
# adapt parentless index to the rest of the code.
1496
for node in self._graph_index.iter_entries(keys):
1497
yield node[0], node[1], node[2], ()
1498
found_keys.add(node[1])
1500
missing_keys = keys.difference(found_keys)
1502
raise RevisionNotPresent(missing_keys.pop(), self)
1504
def _present_keys(self, version_ids):
1506
node[1] for node in self._get_entries(version_ids)])
1508
def _parentless_ancestry(self, versions):
1509
"""Honour the get_ancestry API for parentless knit indices."""
1510
wanted_keys = self._version_ids_to_keys(versions)
1511
present_keys = self._present_keys(wanted_keys)
1512
missing = set(wanted_keys).difference(present_keys)
1514
raise RevisionNotPresent(missing.pop(), self)
1515
return list(self._keys_to_version_ids(present_keys))
1517
def get_ancestry(self, versions, topo_sorted=True):
1518
"""See VersionedFile.get_ancestry."""
1519
if not self._parents:
1520
return self._parentless_ancestry(versions)
1521
# XXX: This will do len(history) index calls - perhaps
1522
# it should be altered to be a index core feature?
1523
# get a graph of all the mentioned versions:
1526
versions = self._version_ids_to_keys(versions)
1527
pending = set(versions)
1529
# get all pending nodes
1530
this_iteration = pending
1531
new_nodes = self._get_entries(this_iteration)
1534
for (index, key, value, node_refs) in new_nodes:
1535
# dont ask for ghosties - otherwise
1536
# we we can end up looping with pending
1537
# being entirely ghosted.
1538
graph[key] = [parent for parent in node_refs[0]
1539
if parent not in ghosts]
1541
for parent in graph[key]:
1542
# dont examine known nodes again
1547
ghosts.update(this_iteration.difference(found))
1548
if versions.difference(graph):
1549
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1551
result_keys = topo_sort(graph.items())
1553
result_keys = graph.iterkeys()
1554
return [key[0] for key in result_keys]
1556
def get_ancestry_with_ghosts(self, versions):
1557
"""See VersionedFile.get_ancestry."""
1558
if not self._parents:
1559
return self._parentless_ancestry(versions)
1560
# XXX: This will do len(history) index calls - perhaps
1561
# it should be altered to be a index core feature?
1562
# get a graph of all the mentioned versions:
1564
versions = self._version_ids_to_keys(versions)
1565
pending = set(versions)
1567
# get all pending nodes
1568
this_iteration = pending
1569
new_nodes = self._get_entries(this_iteration)
1571
for (index, key, value, node_refs) in new_nodes:
1572
graph[key] = node_refs[0]
1574
for parent in graph[key]:
1575
# dont examine known nodes again
1579
missing_versions = this_iteration.difference(graph)
1580
missing_needed = versions.intersection(missing_versions)
1582
raise RevisionNotPresent(missing_needed.pop(), self)
1583
for missing_version in missing_versions:
1584
# add a key, no parents
1585
graph[missing_version] = []
1586
pending.discard(missing_version) # don't look for it
1587
result_keys = topo_sort(graph.items())
1588
return [key[0] for key in result_keys]
1590
def get_graph(self):
1591
"""Return a list of the node:parents lists from this knit index."""
1592
if not self._parents:
1593
return [(key, ()) for key in self.get_versions()]
1595
for index, key, value, refs in self._graph_index.iter_all_entries():
1596
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1599
def iter_parents(self, version_ids):
1600
"""Iterate through the parents for many version ids.
1602
:param version_ids: An iterable yielding version_ids.
1603
:return: An iterator that yields (version_id, parents). Requested
1604
version_ids not present in the versioned file are simply skipped.
1605
The order is undefined, allowing for different optimisations in
1606
the underlying implementation.
1609
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1611
present_parents = set()
1612
for node in all_nodes:
1613
all_parents.update(node[3][0])
1614
# any node we are querying must be present
1615
present_parents.add(node[1])
1616
unknown_parents = all_parents.difference(present_parents)
1617
present_parents.update(self._present_keys(unknown_parents))
1618
for node in all_nodes:
1620
for parent in node[3][0]:
1621
if parent in present_parents:
1622
parents.append(parent[0])
1623
yield node[1][0], tuple(parents)
1625
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1626
yield node[1][0], ()
1628
def num_versions(self):
1629
return len(list(self._graph_index.iter_all_entries()))
1631
__len__ = num_versions
1633
def get_versions(self):
1634
"""Get all the versions in the file. not topologically sorted."""
1635
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1637
def has_version(self, version_id):
1638
"""True if the version is in the index."""
1639
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1641
def _keys_to_version_ids(self, keys):
1642
return tuple(key[0] for key in keys)
1644
def get_position(self, version_id):
1645
"""Return details needed to access the version.
1647
:return: a tuple (index, data position, size) to hand to the access
1648
logic to get the record.
1650
node = self._get_node(version_id)
1651
bits = node[2][1:].split(' ')
1652
return node[0], int(bits[0]), int(bits[1])
1654
def get_method(self, version_id):
1655
"""Return compression method of specified version."""
1656
if not self._deltas:
1658
return self._parent_compression(self._get_node(version_id)[3][1])
1660
def _parent_compression(self, reference_list):
1661
# use the second reference list to decide if this is delta'd or not.
1662
if len(reference_list):
1667
def _get_node(self, version_id):
1669
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1671
raise RevisionNotPresent(version_id, self)
1673
def get_options(self, version_id):
1674
"""Return a string represention options.
1678
node = self._get_node(version_id)
1679
if not self._deltas:
1680
options = ['fulltext']
1682
options = [self._parent_compression(node[3][1])]
1683
if node[2][0] == 'N':
1684
options.append('no-eol')
1687
def get_parents(self, version_id):
1688
"""Return parents of specified version ignoring ghosts."""
1689
parents = list(self.iter_parents([version_id]))
1692
raise errors.RevisionNotPresent(version_id, self)
1693
return parents[0][1]
1695
def get_parents_with_ghosts(self, version_id):
1696
"""Return parents of specified version with ghosts."""
1697
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1698
check_present=True))
1699
if not self._parents:
1701
return self._keys_to_version_ids(nodes[0][3][0])
1703
def check_versions_present(self, version_ids):
1704
"""Check that all specified versions are present."""
1705
keys = self._version_ids_to_keys(version_ids)
1706
present = self._present_keys(keys)
1707
missing = keys.difference(present)
1709
raise RevisionNotPresent(missing.pop(), self)
1711
def add_version(self, version_id, options, access_memo, parents):
1712
"""Add a version record to the index."""
1713
return self.add_versions(((version_id, options, access_memo, parents),))
1715
def add_versions(self, versions):
1716
"""Add multiple versions to the index.
1718
This function does not insert data into the Immutable GraphIndex
1719
backing the KnitGraphIndex, instead it prepares data for insertion by
1720
the caller and checks that it is safe to insert then calls
1721
self._add_callback with the prepared GraphIndex nodes.
1723
:param versions: a list of tuples:
1724
(version_id, options, pos, size, parents).
1726
if not self._add_callback:
1727
raise errors.ReadOnlyError(self)
1728
# we hope there are no repositories with inconsistent parentage
1733
for (version_id, options, access_memo, parents) in versions:
1734
index, pos, size = access_memo
1735
key = (version_id, )
1736
parents = tuple((parent, ) for parent in parents)
1737
if 'no-eol' in options:
1741
value += "%d %d" % (pos, size)
1742
if not self._deltas:
1743
if 'line-delta' in options:
1744
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1747
if 'line-delta' in options:
1748
node_refs = (parents, (parents[0],))
1750
node_refs = (parents, ())
1752
node_refs = (parents, )
1755
raise KnitCorrupt(self, "attempt to add node with parents "
1756
"in parentless index.")
1758
keys[key] = (value, node_refs)
1759
present_nodes = self._get_entries(keys)
1760
for (index, key, value, node_refs) in present_nodes:
1761
if (value, node_refs) != keys[key]:
1762
raise KnitCorrupt(self, "inconsistent details in add_versions"
1763
": %s %s" % ((value, node_refs), keys[key]))
1767
for key, (value, node_refs) in keys.iteritems():
1768
result.append((key, value, node_refs))
1770
for key, (value, node_refs) in keys.iteritems():
1771
result.append((key, value))
1772
self._add_callback(result)
1774
def _version_ids_to_keys(self, version_ids):
1775
return set((version_id, ) for version_id in version_ids)
1778
class _KnitAccess(object):
1779
"""Access to knit records in a .knit file."""
1781
def __init__(self, transport, filename, _file_mode, _dir_mode,
1782
_need_to_create, _create_parent_dir):
1783
"""Create a _KnitAccess for accessing and inserting data.
1785
:param transport: The transport the .knit is located on.
1786
:param filename: The filename of the .knit.
1788
self._transport = transport
1789
self._filename = filename
1790
self._file_mode = _file_mode
1791
self._dir_mode = _dir_mode
1792
self._need_to_create = _need_to_create
1793
self._create_parent_dir = _create_parent_dir
1795
def add_raw_records(self, sizes, raw_data):
1796
"""Add raw knit bytes to a storage area.
1798
The data is spooled to whereever the access method is storing data.
1800
:param sizes: An iterable containing the size of each raw data segment.
1801
:param raw_data: A bytestring containing the data.
1802
:return: A list of memos to retrieve the record later. Each memo is a
1803
tuple - (index, pos, length), where the index field is always None
1804
for the .knit access method.
1806
assert type(raw_data) == str, \
1807
'data must be plain bytes was %s' % type(raw_data)
1808
if not self._need_to_create:
1809
base = self._transport.append_bytes(self._filename, raw_data)
1811
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1812
create_parent_dir=self._create_parent_dir,
1813
mode=self._file_mode,
1814
dir_mode=self._dir_mode)
1815
self._need_to_create = False
1819
result.append((None, base, size))
1824
"""IFF this data access has its own storage area, initialise it.
1828
self._transport.put_bytes_non_atomic(self._filename, '',
1829
mode=self._file_mode)
1831
def open_file(self):
1832
"""IFF this data access can be represented as a single file, open it.
1834
For knits that are not mapped to a single file on disk this will
1837
:return: None or a file handle.
1840
return self._transport.get(self._filename)
1845
def get_raw_records(self, memos_for_retrieval):
1846
"""Get the raw bytes for a records.
1848
:param memos_for_retrieval: An iterable containing the (index, pos,
1849
length) memo for retrieving the bytes. The .knit method ignores
1850
the index as there is always only a single file.
1851
:return: An iterator over the bytes of the records.
1853
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1854
for pos, data in self._transport.readv(self._filename, read_vector):
1858
class _PackAccess(object):
1859
"""Access to knit records via a collection of packs."""
1861
def __init__(self, index_to_packs, writer=None):
1862
"""Create a _PackAccess object.
1864
:param index_to_packs: A dict mapping index objects to the transport
1865
and file names for obtaining data.
1866
:param writer: A tuple (pack.ContainerWriter, write_index) which
1867
contains the pack to write, and the index that reads from it will
1871
self.container_writer = writer[0]
1872
self.write_index = writer[1]
1874
self.container_writer = None
1875
self.write_index = None
1876
self.indices = index_to_packs
1878
def add_raw_records(self, sizes, raw_data):
1879
"""Add raw knit bytes to a storage area.
1881
The data is spooled to the container writer in one bytes-record per
1884
:param sizes: An iterable containing the size of each raw data segment.
1885
:param raw_data: A bytestring containing the data.
1886
:return: A list of memos to retrieve the record later. Each memo is a
1887
tuple - (index, pos, length), where the index field is the
1888
write_index object supplied to the PackAccess object.
1890
assert type(raw_data) == str, \
1891
'data must be plain bytes was %s' % type(raw_data)
1895
p_offset, p_length = self.container_writer.add_bytes_record(
1896
raw_data[offset:offset+size], [])
1898
result.append((self.write_index, p_offset, p_length))
1902
"""Pack based knits do not get individually created."""
1904
def get_raw_records(self, memos_for_retrieval):
1905
"""Get the raw bytes for a records.
1907
:param memos_for_retrieval: An iterable containing the (index, pos,
1908
length) memo for retrieving the bytes. The Pack access method
1909
looks up the pack to use for a given record in its index_to_pack
1911
:return: An iterator over the bytes of the records.
1913
# first pass, group into same-index requests
1915
current_index = None
1916
for (index, offset, length) in memos_for_retrieval:
1917
if current_index == index:
1918
current_list.append((offset, length))
1920
if current_index is not None:
1921
request_lists.append((current_index, current_list))
1922
current_index = index
1923
current_list = [(offset, length)]
1924
# handle the last entry
1925
if current_index is not None:
1926
request_lists.append((current_index, current_list))
1927
for index, offsets in request_lists:
1928
transport, path = self.indices[index]
1929
reader = pack.make_readv_reader(transport, path, offsets)
1930
for names, read_func in reader.iter_records():
1931
yield read_func(None)
1933
def open_file(self):
1934
"""Pack based knits have no single file."""
1937
def set_writer(self, writer, index, (transport, packname)):
1938
"""Set a writer to use for adding data."""
1939
self.indices[index] = (transport, packname)
1940
self.container_writer = writer
1941
self.write_index = index
1944
class _KnitData(object):
1945
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1947
The KnitData class provides the logic for parsing and using knit records,
1948
making use of an access method for the low level read and write operations.
1951
def __init__(self, access):
1952
"""Create a KnitData object.
1954
:param access: The access method to use. Access methods such as
1955
_KnitAccess manage the insertion of raw records and the subsequent
1956
retrieval of the same.
1958
self._access = access
1959
self._checked = False
1960
# TODO: jam 20060713 conceptually, this could spill to disk
1961
# if the cached size gets larger than a certain amount
1962
# but it complicates the model a bit, so for now just use
1963
# a simple dictionary
1965
self._do_cache = False
1967
def enable_cache(self):
1968
"""Enable caching of reads."""
1969
self._do_cache = True
1971
def clear_cache(self):
1972
"""Clear the record cache."""
1973
self._do_cache = False
1976
def _open_file(self):
1977
return self._access.open_file()
1979
def _record_to_data(self, version_id, digest, lines):
1980
"""Convert version_id, digest, lines into a raw data block.
1982
:return: (len, a StringIO instance with the raw data ready to read.)
1984
bytes = (''.join(chain(
1985
["version %s %d %s\n" % (version_id,
1989
["end %s\n" % version_id])))
1990
assert bytes.__class__ == str
1991
compressed_bytes = bytes_to_gzip(bytes)
1992
return len(compressed_bytes), compressed_bytes
1994
def add_raw_records(self, sizes, raw_data):
1995
"""Append a prepared record to the data file.
1997
:param sizes: An iterable containing the size of each raw data segment.
1998
:param raw_data: A bytestring containing the data.
1999
:return: a list of index data for the way the data was stored.
2000
See the access method add_raw_records documentation for more
2003
return self._access.add_raw_records(sizes, raw_data)
2005
def add_record(self, version_id, digest, lines):
2006
"""Write new text record to disk.
2008
Returns index data for retrieving it later, as per add_raw_records.
2010
size, bytes = self._record_to_data(version_id, digest, lines)
2011
result = self.add_raw_records([size], bytes)
2013
self._cache[version_id] = bytes
2016
def _parse_record_header(self, version_id, raw_data):
2017
"""Parse a record header for consistency.
2019
:return: the header and the decompressor stream.
2020
as (stream, header_record)
2022
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2024
rec = self._check_header(version_id, df.readline())
2025
except Exception, e:
2026
raise KnitCorrupt(self._access,
2027
"While reading {%s} got %s(%s)"
2028
% (version_id, e.__class__.__name__, str(e)))
2031
def _check_header(self, version_id, line):
2034
raise KnitCorrupt(self._access,
2035
'unexpected number of elements in record header')
2036
if rec[1] != version_id:
2037
raise KnitCorrupt(self._access,
2038
'unexpected version, wanted %r, got %r'
2039
% (version_id, rec[1]))
2042
def _parse_record(self, version_id, data):
2044
# 4168 calls in 2880 217 internal
2045
# 4168 calls to _parse_record_header in 2121
2046
# 4168 calls to readlines in 330
2047
df = GzipFile(mode='rb', fileobj=StringIO(data))
2050
record_contents = df.readlines()
2051
except Exception, e:
2052
raise KnitCorrupt(self._access,
2053
"While reading {%s} got %s(%s)"
2054
% (version_id, e.__class__.__name__, str(e)))
2055
header = record_contents.pop(0)
2056
rec = self._check_header(version_id, header)
2058
last_line = record_contents.pop()
2059
if len(record_contents) != int(rec[2]):
2060
raise KnitCorrupt(self._access,
2061
'incorrect number of lines %s != %s'
2063
% (len(record_contents), int(rec[2]),
2065
if last_line != 'end %s\n' % rec[1]:
2066
raise KnitCorrupt(self._access,
2067
'unexpected version end line %r, wanted %r'
2068
% (last_line, version_id))
2070
return record_contents, rec[3]
2072
def read_records_iter_raw(self, records):
2073
"""Read text records from data file and yield raw data.
2075
This unpacks enough of the text record to validate the id is
2076
as expected but thats all.
2078
# setup an iterator of the external records:
2079
# uses readv so nice and fast we hope.
2081
# grab the disk data needed.
2083
# Don't check _cache if it is empty
2084
needed_offsets = [index_memo for version_id, index_memo
2086
if version_id not in self._cache]
2088
needed_offsets = [index_memo for version_id, index_memo
2091
raw_records = self._access.get_raw_records(needed_offsets)
2093
for version_id, index_memo in records:
2094
if version_id in self._cache:
2095
# This data has already been validated
2096
data = self._cache[version_id]
2098
data = raw_records.next()
2100
self._cache[version_id] = data
2102
# validate the header
2103
df, rec = self._parse_record_header(version_id, data)
2105
yield version_id, data
2107
def read_records_iter(self, records):
2108
"""Read text records from data file and yield result.
2110
The result will be returned in whatever is the fastest to read.
2111
Not by the order requested. Also, multiple requests for the same
2112
record will only yield 1 response.
2113
:param records: A list of (version_id, pos, len) entries
2114
:return: Yields (version_id, contents, digest) in the order
2115
read, not the order requested
2121
# Skip records we have alread seen
2122
yielded_records = set()
2123
needed_records = set()
2124
for record in records:
2125
if record[0] in self._cache:
2126
if record[0] in yielded_records:
2128
yielded_records.add(record[0])
2129
data = self._cache[record[0]]
2130
content, digest = self._parse_record(record[0], data)
2131
yield (record[0], content, digest)
2133
needed_records.add(record)
2134
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2136
needed_records = sorted(set(records), key=operator.itemgetter(1))
2138
if not needed_records:
2141
# The transport optimizes the fetching as well
2142
# (ie, reads continuous ranges.)
2143
raw_data = self._access.get_raw_records(
2144
[index_memo for version_id, index_memo in needed_records])
2146
for (version_id, index_memo), data in \
2147
izip(iter(needed_records), raw_data):
2148
content, digest = self._parse_record(version_id, data)
2150
self._cache[version_id] = data
2151
yield version_id, content, digest
2153
def read_records(self, records):
2154
"""Read records into a dictionary."""
2156
for record_id, content, digest in \
2157
self.read_records_iter(records):
2158
components[record_id] = (content, digest)
2162
class InterKnit(InterVersionedFile):
2163
"""Optimised code paths for knit to knit operations."""
2165
_matching_file_from_factory = KnitVersionedFile
2166
_matching_file_to_factory = KnitVersionedFile
2169
def is_compatible(source, target):
2170
"""Be compatible with knits. """
2172
return (isinstance(source, KnitVersionedFile) and
2173
isinstance(target, KnitVersionedFile))
2174
except AttributeError:
2177
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2178
"""See InterVersionedFile.join."""
2179
assert isinstance(self.source, KnitVersionedFile)
2180
assert isinstance(self.target, KnitVersionedFile)
2182
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2187
pb = ui.ui_factory.nested_progress_bar()
2189
version_ids = list(version_ids)
2190
if None in version_ids:
2191
version_ids.remove(None)
2193
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2194
this_versions = set(self.target._index.get_versions())
2195
needed_versions = self.source_ancestry - this_versions
2196
cross_check_versions = self.source_ancestry.intersection(this_versions)
2197
mismatched_versions = set()
2198
for version in cross_check_versions:
2199
# scan to include needed parents.
2200
n1 = set(self.target.get_parents_with_ghosts(version))
2201
n2 = set(self.source.get_parents_with_ghosts(version))
2203
# FIXME TEST this check for cycles being introduced works
2204
# the logic is we have a cycle if in our graph we are an
2205
# ancestor of any of the n2 revisions.
2211
parent_ancestors = self.source.get_ancestry(parent)
2212
if version in parent_ancestors:
2213
raise errors.GraphCycleError([parent, version])
2214
# ensure this parent will be available later.
2215
new_parents = n2.difference(n1)
2216
needed_versions.update(new_parents.difference(this_versions))
2217
mismatched_versions.add(version)
2219
if not needed_versions and not mismatched_versions:
2221
full_list = topo_sort(self.source.get_graph())
2223
version_list = [i for i in full_list if (not self.target.has_version(i)
2224
and i in needed_versions)]
2228
copy_queue_records = []
2230
for version_id in version_list:
2231
options = self.source._index.get_options(version_id)
2232
parents = self.source._index.get_parents_with_ghosts(version_id)
2233
# check that its will be a consistent copy:
2234
for parent in parents:
2235
# if source has the parent, we must :
2236
# * already have it or
2237
# * have it scheduled already
2238
# otherwise we don't care
2239
assert (self.target.has_version(parent) or
2240
parent in copy_set or
2241
not self.source.has_version(parent))
2242
index_memo = self.source._index.get_position(version_id)
2243
copy_queue_records.append((version_id, index_memo))
2244
copy_queue.append((version_id, options, parents))
2245
copy_set.add(version_id)
2247
# data suck the join:
2249
total = len(version_list)
2252
for (version_id, raw_data), \
2253
(version_id2, options, parents) in \
2254
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2256
assert version_id == version_id2, 'logic error, inconsistent results'
2258
pb.update("Joining knit", count, total)
2259
raw_records.append((version_id, options, parents, len(raw_data)))
2260
raw_datum.append(raw_data)
2261
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2263
for version in mismatched_versions:
2264
# FIXME RBC 20060309 is this needed?
2265
n1 = set(self.target.get_parents_with_ghosts(version))
2266
n2 = set(self.source.get_parents_with_ghosts(version))
2267
# write a combined record to our history preserving the current
2268
# parents as first in the list
2269
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2270
self.target.fix_parents(version, new_parents)
2276
InterVersionedFile.register_optimiser(InterKnit)
2279
class WeaveToKnit(InterVersionedFile):
2280
"""Optimised code paths for weave to knit operations."""
2282
_matching_file_from_factory = bzrlib.weave.WeaveFile
2283
_matching_file_to_factory = KnitVersionedFile
2286
def is_compatible(source, target):
2287
"""Be compatible with weaves to knits."""
2289
return (isinstance(source, bzrlib.weave.Weave) and
2290
isinstance(target, KnitVersionedFile))
2291
except AttributeError:
2294
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2295
"""See InterVersionedFile.join."""
2296
assert isinstance(self.source, bzrlib.weave.Weave)
2297
assert isinstance(self.target, KnitVersionedFile)
2299
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2304
pb = ui.ui_factory.nested_progress_bar()
2306
version_ids = list(version_ids)
2308
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2309
this_versions = set(self.target._index.get_versions())
2310
needed_versions = self.source_ancestry - this_versions
2311
cross_check_versions = self.source_ancestry.intersection(this_versions)
2312
mismatched_versions = set()
2313
for version in cross_check_versions:
2314
# scan to include needed parents.
2315
n1 = set(self.target.get_parents_with_ghosts(version))
2316
n2 = set(self.source.get_parents(version))
2317
# if all of n2's parents are in n1, then its fine.
2318
if n2.difference(n1):
2319
# FIXME TEST this check for cycles being introduced works
2320
# the logic is we have a cycle if in our graph we are an
2321
# ancestor of any of the n2 revisions.
2327
parent_ancestors = self.source.get_ancestry(parent)
2328
if version in parent_ancestors:
2329
raise errors.GraphCycleError([parent, version])
2330
# ensure this parent will be available later.
2331
new_parents = n2.difference(n1)
2332
needed_versions.update(new_parents.difference(this_versions))
2333
mismatched_versions.add(version)
2335
if not needed_versions and not mismatched_versions:
2337
full_list = topo_sort(self.source.get_graph())
2339
version_list = [i for i in full_list if (not self.target.has_version(i)
2340
and i in needed_versions)]
2344
total = len(version_list)
2345
for version_id in version_list:
2346
pb.update("Converting to knit", count, total)
2347
parents = self.source.get_parents(version_id)
2348
# check that its will be a consistent copy:
2349
for parent in parents:
2350
# if source has the parent, we must already have it
2351
assert (self.target.has_version(parent))
2352
self.target.add_lines(
2353
version_id, parents, self.source.get_lines(version_id))
2356
for version in mismatched_versions:
2357
# FIXME RBC 20060309 is this needed?
2358
n1 = set(self.target.get_parents_with_ghosts(version))
2359
n2 = set(self.source.get_parents(version))
2360
# write a combined record to our history preserving the current
2361
# parents as first in the list
2362
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2363
self.target.fix_parents(version, new_parents)
2369
InterVersionedFile.register_optimiser(WeaveToKnit)
2372
# Deprecated, use PatienceSequenceMatcher instead
2373
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2376
def annotate_knit(knit, revision_id):
2377
"""Annotate a knit with no cached annotations.
2379
This implementation is for knits with no cached annotations.
2380
It will work for knits with cached annotations, but this is not
2383
ancestry = knit.get_ancestry(revision_id)
2384
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2386
for candidate in ancestry:
2387
if candidate in annotations:
2389
parents = knit.get_parents(candidate)
2390
if len(parents) == 0:
2392
elif knit._index.get_method(candidate) != 'line-delta':
2395
parent, sha1, noeol, delta = knit.get_delta(candidate)
2396
blocks = KnitContent.get_line_delta_blocks(delta,
2397
fulltext[parents[0]], fulltext[candidate])
2398
annotations[candidate] = list(annotate.reannotate([annotations[p]
2399
for p in parents], fulltext[candidate], candidate, blocks))
2400
return iter(annotations[revision_id])
2404
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2406
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
